Research On Service Efficiency Model Of Shipboard Public Computing Environment

In recent years,with the improvement of the strategic position of coastal defense and the development of new naval warfare mode,shipborne applications tend to be complex and intelligent.With the development of Shipborne applications,on the one hand,it is difficult to increase the computing efficiency by unlimited resources in resource constrained shipborne environment,and the complex dynamic characteristics of battlefield put forward higher requirements for resource optimization control.On the other hand,due to the particularity of shipboard service object,it also challenges the computing service.For example,it can not meet the time constraints of strong real-time tasks on computing services,and the existing passive resource failure mode leads to low efficiency problems.This paper focuses on the service quality assurance and optimization of shipboard public computing environment.Aiming at the high-quality requirements of computing services,the existing shipborne public computing environment architecture lacks complete design elements of service quality assurance.Therefore,it is difficult to guarantee high-quality computing services in the face of increasingly complex modern combat mode.Moreover,the existing shipboard public computing environment lacks the interface of service quality prediction and tuning control.Therefore,it is difficult to achieve the quantitative evaluation and optimization of service quality actively,and only rely on the passive post analysis and qualitative evaluation,can not make a positive contribution to the guarantee of service quality.1)Service performance prediction and tuning control interface are integrated into the architecture,and the resource health management interface is integrated into the architecture.Moreover,in order to meet the time constraints of Shipborne strong real-time tasks on computing services,the architecture is optimized towards real-time by actively customizing the optimization design scheme.Finally,through the design of service-oriented real-time computing graph partition optimization algorithm,the optimization design idea of the architecture is verified.The experimental data show that the real-time response performance of service-oriented real-time computing graph partition algorithm is up to 900 ms,which is about 80% higher than the slowest random partition algorithm,and about 67% higher than the fastest CP partition algorithm,making the architecture more real-time.2)In the shipboard environment with limited resources,the BP neural network optimization algorithm with certain accuracy and efficiency is customized.BP neural network algorithm is optimized comprehensively from data allocation,parallel computing model and gradient descent algorithm.The data partition strategy based on computing power allocation,grouping fine-grained inter layer parallel algorithm,and two-stage gradient descent algorithm are designed to provide the support of modeling algorithm level for the establishment of efficient and accurate service efficiency model.The experimental results show that the service-oriented BP neural network optimization algorithm has a great improvement in the computational efficiency,the average speedup is about 50%,and the maximum speedup is 54%.At the same time,the average accuracy of top-5 is about 70%.Therefore,the customized BP neural network optimization algorithm provides the support of modeling algorithm level for the service effectiveness model.3)The intelligent prediction model of two-level service efficiency includes the overall service efficiency prediction model and the virtual machine service efficiency prediction model.In order to build an effective service efficiency model,the input characteristics of the service efficiency model are extracted from the system and virtual machine levels respectively,and the output indexes of the service efficiency model are defined.Specifically,three tuples(availability,service capability,service efficiency)are used to represent the overall service efficiency,and two tuples(service capability,service efficiency)are used to represent the virtual machine service efficiency.Based on the defined input characteristics and service efficiency output indicators,the BP neural network optimization model is used to complete the construction of two-level service efficiency model,and the accuracy of service efficiency model prediction is evaluated through three kinds of errors.The experimental results show that,compared with the linear regression model,the BP neural network optimization algorithm improves the top-5 accuracy of the overall service efficiency prediction by about 20%,and the top-5 accuracy of the virtual machine service efficiency prediction by about 22%.Compared with the BP neural network algorithm without task optimization strategy,the top-5 accuracy of the overall service efficiency prediction is improved by about 16%,and the top-5accuracy of the virtual machine service efficiency prediction is improved by about12%.4)In order to maintain a high level of service efficiency,the optimization control strategy is constructed based on the designed two-stage neural network inverse model.In addition,in the face of limited shipboard resources and environment,a two-level service efficiency optimization model based on priority is designed to balance the service efficiency optimization target demand of system level and user level.The experimental results show that the average absolute error,mean square error and root mean square error of the two-level service efficiency inverse model are the smallest compared with the inverse model constructed by genetic algorithm and BP neural network.The average absolute error of 14 components is 0.029,the average absolute error of genetic algorithm is 0.25,and the average absolute error of unoptimized BP neural network is 0.57.